conservation, and exon sequence homology
Holger Pillmann*, Klas Hatje*, Florian Odronitz, Björn Hammesfahr, and Martin Kollmar1§
Abteilung NMR basierte Strukturbiologie, Max-Planck-Institut für Biophysikalische Chemie, Am Fassberg 11, D-37077 Göttingen, Germany
* These authors contributed equally to the work.
§ Corresponding author
BMC Bioinformatics
Published: 30 June 2011
BMC Bioinformatics 2011 12:270 doi:10.1186/1471-2105-12-270 This article is available from http://www.biomedcentral.com/1471-2105/12/270
2.5.1 Abstract
Background
Alternative splicing of pre-mature RNA is an important process eukaryotes utilize to increase their repertoire of different protein products. Several types of different alternative splice forms exist including exon skipping, differential splicing of exons at their 3'- or 5'-end, intron retention, and mutually exclusive splicing. The latter term is used for clusters of internal exons that are spliced in a mutually exclusive manner.
Results
We have implemented an extension to the WebScipio software to search for mutually exclusive exons. Here, the search is based on the precondition that mutually exclusive exons encode regions of the same structural part of the protein product. This precondition provides restrictions to the search for candidate exons concerning their length, splice site conservation and reading frame preservation, and overall homology. Mutually exclusive exons that are not homologous and not of about the same length will not be found. Using the new algorithm, mutually exclusive exons in several example genes, a dynein heavy chain, a muscle myosin heavy chain, and Dscam were correctly identified. In addition, the algorithm was applied to the whole Drosophila melanogaster X chromosome and the results were compared to the Flybase annotation and an ab initio prediction. Clusters of mutually exclusive exons might be subsequent to each other and might encode dozens of exons.
Conclusions
This is the first implementation of an automatic search for mutually exclusive exons in eukaryotes. Exons are predicted and reconstructed in the same run providing the complete gene structure for the protein query of interest. WebScipio offers high quality gene structure figures with the clusters of mutually exclusive exons colour-coded, and several analysis tools for further manual inspection. The genome scale analysis of all genes of the Drosophila melanogaster X chromosome showed that WebScipio is able to find all but two of the 28 annotated mutually exclusive spliced exons and predicts 39 new candidate exons. Thus, WebScipio should be able to identify mutually exclusive spliced exons in any query sequence from any species with a very high probability. WebScipio is freely available to academics at http://www.webscipio.org.
2.5.2 Background
Eukaryotes can enhance their repertoire of different protein products by alternative splicing of the corresponding genes (307). Since the first description of alternative splicing of precursor mRNA almost 30 years ago (308,309) the suggested and verified percentage of human genes that are spliced into alternative transcripts has steadily risen (for reviews see for example (22,310)). Very recently, two studies using high-throughput sequencing indicate that every single human gene containing more than one exon is transcribed and processed to yield multiple mRNAs (15,311).
Mainly, five different types of alternative splicing affect the resulting translated protein product (14,312,313): The first type is exon skipping, in which an exon, also called cassette exon, is spliced out of the transcript together with its flanking introns. The second and third types are the alternative splicing of the 3' splice site and 5' splice site, respectively. Here, two or more splice sites are recognized at one end of the exon. The fourth type is intron retention in which part of an exon is either spliced (like a regular intron) or retained in the mature mRNA transcript. While exon skipping and alternative 3' splice site selection account for most alternative splicing events in higher eukaryotes (16,17), the most prevalent type of alternative splicing in plants, fungi, and protozoa is intron retention (18). The fifths type is called mutually exclusive splicing and is used for clusters of internal exons that are spliced in a mutually exclusive manner. It is important to note that the term mutually exclusive splicing is only used for these specific clusters of exons. Mutually exclusive splicing demands a specific mechanism for the regulated splicing of exactly one of the exons of such a cluster. Recent analyses have shown that this mechanism might be based on intra-intronic RNA pairings that are conserved at the secondary structure level (314–316). These alternatively spliced exons must not be mixed up with exons that seem to be spliced in a mutually exclusive manner based on their annotation. This especially accounts for terminal exons that are alternatively spliced in
conjunction with the use of alternative promoters or 3’-end processing sites (for a review see for example (317)). The regulation of the splicing of these types need not be at the level of splicing.
To our knowledge, the only study to identify and predict regions in silico that might contain mutually exclusive spliced exons used a method of local similarity of genomic regions at the nucleotide level (318). Assuming that clusters of mutually exclusive exons evolved by one or several rounds of single-exon duplications, given gene locations were self-aligned using a pairwise local alignment algorithm to derive similar regions. Those regions were regarded as candidate regions, and mutually exclusive exons were only predicted by verification through EST and cDNA data. The method itself cannot determine exons including intron splice sites, and is not able to identify mutually exclusive exons whose DNA sequences have diverged considerably. False positive candidates are detected in regions that contain clusters of duplicated genes, and in regions containing pseudo-exons (e.g. pseudo-exons that are in the process of being lost containing frame-shifts and in-frame stop codons, and missing correct splice sites).
Here, we propose a different approach that is based on the knowledge of creating meaningful transcripts. We presume that most mutually exclusive exons encode the same region of the resulting protein structure. These regions are embedded in the surrounding three-dimensional structure and thus alternative exons must preserve all structurally important contacts between the corresponding local structure elements. A demonstrative example is the alternatively spliced motor domain of the muscle myosin heavy chain in arthropods (24). In Drosophila, four clusters of mutually exclusive spliced exons encode regions of the motor domain, and the variability of creating different transcripts and further fine-tune the motor domain function is even enhanced in the waterflea Daphnia magna by four additional clusters. One of the clusters contains exons encoding the so-called relay helix and subsequent relay loop, a structural element that starts at switch-2 embedded in the middle of the motor domain and ends at the connection to the converter domain. This whole relay element converts small conformational changes at the ATP-binding site to large movements of the lever arm (32). Retaining structural integrity is therefore indispensable for mutually exclusive exons. Of course, parts of the exons might also encode loop regions, but also those parts must at least partly be conserved to retain their general function.
Based on these preconditions we apply the following constrains to our search for mutually exclusive exons: A) Mutually exclusive exons must have about the same length (allowing some length difference for e.g. parts encoding loop regions). B) They must have conserved splice site patterns (e.g. a GT 5’ intron splice site cannot be combined with an AC 3’ splice site) and the reading frame of the exon must be conserved. C) They must show sequence
similarity. These features have been implemented in an extension to the WebScipio software. The application of the algorithm to various genes from several eukaryotes, and to all genes of the X chromosome of Drosophila melanogaster is demonstrated.
2.5.3 Methods
The search algorithm has been implemented as an extension to the WebScipio web application (13). It is based on the exon-intron gene structure reconstructed by Scipio (212). The extension is written in the Ruby programming language (78) and fully integrated into WebScipio to facilitate user interaction, and visualization and analysis of the results. WebScipio uses the web framework Ruby on Rails (79). To make the session storage fast, flexible, and scalable a database backend consisting of Tokyo Cabinet and Tokyo Tyrant (293) is used. To run jobs in background the Rails plug-in Workling in combination with Spawn (291,292) is applied.
Figure 2.5-1: Activity diagram of the search algorithm. The activity diagram shows the processing steps of the search algorithm and the influence of the parameters on each step. The run starts with an exon-intron gene structure determined by Scipio. Based on the chosen parameters the exons and corresponding introns are selected and searched for mutually exclusive spliced exon candidates. The candidates are processed and filtered. These steps are repeated in the case of a recursive run. In the end, the algorithm outputs the exon-intron structure including mutually exclusive spliced exons.
Search algorithm
The new algorithm divides into several steps, which are executed for each original exon (Figure 2.5-1, a detailed activity diagram is available as Additional file 2.5.9.1). It assumes that mutually exclusive spliced exons share the following features: Firstly, mutually exclusive spliced exons have a similar length; secondly, their splice sites and reading frames are conserved; thirdly, they are homologous.
For each internal exon ("original exon") the two surrounding introns (or optionally all introns of the gene) are scanned for exon candidates that have a similar length. These exon candidates must introduce introns with the following splice site pattern: GT---AG, GC---AG, GG---GC---AG, and AT---AC. Firstly, the algorithm looks for the nucleotide pairs AG or AC in the intron sequence, which define start sites of exon candidates and 3’ splice sites of the proposed intron. If the intron in front of the original exon starts with GT, GC or GG the algorithm searches for AG, if it starts with an AT the algorithm searches for AC. Secondly, the algorithm looks for the nucleotide pairs GT, GC, GG and AT in the intron sequence, which define ends of exon candidates and 5’ splice sites of the proposed intron. If the intron following the original exon ends with AG the algorithm searches for GT, GC and GG, if it ends with AC the algorithm searches for AT. The nucleotide sequences between two possible 3’ and 5’ splice sites of the scanned intron that have a length similar to the length of the original exon are considered as exon candidates. The maximum length difference between an exon and its candidate can be adjusted by the allowed length difference parameter in number of amino acids. The default value of this parameter is 20aa.
For terminal exons, the algorithm is able to scan the up- and downstream regions of the gene for exon candidates. The first exon of a protein-coding gene has to start with the start codon ATG. Thus, for the first exon, alternative candidates must start with ATG instead of sharing a theoretical splice site pattern with the first exon. The last exon is followed by a stop codon (TAG, TAA, or TGA) and all exon candidates must be followed by a stop codon instead of sharing a splice site pattern with the last exon. The use of the start codon and stop codon instead of the splice sites can be adjusted by the search with start codon for first exon and search with stop codon for last exon parameters. For example it would be useful to release this restriction in the case where the algorithm searches for alternative exons in a protein fragment. The default of these parameters is to search with a start codon if the first amino acid of the user-provided protein query sequence starts with methionine, and to search with stop codons if the last exon is followed by a stop codon. To reduce the number of candidates it is possible to set the minimal exon length parameter. Original exons, which are shorter than this length, are not considered in the candidate search. The default value for this parameter is 15aa.
The nucleotide sequences of the exon candidates are translated into amino acid sequences using the BioRuby library (260). The candidates are translated in the same reading frame as the original exon, because their nucleotide sequences appear mutually exclusive in the resulting mRNA and thus share the same reading frame. If the translation results in an in-frame stop codon, the candidate is rejected.
Each candidate sequence is aligned to the original exon sequence. If the alignment score is high, the probability that the two exons are homologous is high as well. The optimal global alignment of the two amino acid sequences is calculated with the Gotoh algorithm, which extends the Needleman-Wunsch algorithm by affine gap costs (319,320). For this task, the pair_align program of the SeqAn package (321) is used. The gap penalties are set to -10 for initial gaps and -2 for extending gaps. The Blosum62 matrix is used as substitution matrix (322,323). Because of differences in length and amino acid composition of the clusters of mutually exclusive exons the resulting global alignment scores are not directly comparable.
To normalise the alignment scores each score is divided by the score of the alignment of the original exon sequence to itself. This relative score shows the similarity of the two sequences on a scale from zero to one. Candidates, which have a low alignment score, are rejected. The threshold for rejection can be adjusted in per cent by the minimal score for exons parameter (default: 15%). If candidate regions overlap the highest scoring candidates are retained or, if scores are identical, the longest candidates.
An optional recursive search was implemented to find less similar alternative exons. If this option is selected, the search is repeated with the found alternatively spliced exons as query exons. The number of recursive runs can be adjusted with the maximal recursion depth parameter up to three rounds of recursion (default: recursive search disabled).
Figure 2.5-2: Gene structure representation and detailed alignment view. The figure shows the WebScipio gene structure representation of the Drosophila melanogaster Dscam gene with mutually exclusive spliced exons and a section of the alignment view including exon 5 and the first two identified alternative exon candidates. The colours in the gene structure figure are the same as the colours of the exon identifiers in the text alignment. The opacity of the colours of each alternative exon corresponds to the alignment score of the alternative exon to the original one. This score is shown in the detailed alignment view next to the exon identifier. For each exon the genomic sequence, its translation, and the translation of the original exon is shown. Identical residues are illustrated as dashes and mismatches as red highlighted crosses. The crosses are highlighted in light red for amino acids, which are chemically similar. Gaps are marked as green hyphens.
WebScipio integration
The WebScipio tool allows reconstructing an exon-intron gene structure based on a protein sequence query. This reconstruction step is the basis for the mutually exclusive spliced exon search. The user can enable the search and adjust several parameters in the Advanced Options section of WebScipio. The search will run subsequently to the gene structure reconstruction step. In addition, the user can enable the search after uploading a previously calculated and downloaded Scipio result.
The result of the search is displayed in the Result section of the WebScipio interface (Figure 2.5-2, top). The standard gene structure picture is extended by the predicted mutually exclusive spliced exons. The alternative exons corresponding to the same original exon constitute a cluster. Exons of a cluster get the same colour. The original exon is dark coloured and the corresponding predicted ones are lighter coloured depending on their similarity with respect to the original exon. In the Statistics section the number of exons in each cluster is shown in colour.
The Alignment view (Figure 2.5-2, bottom) offers a detailed analysis at the sequence level.
For each alternative exon the genomic sequence, its translation, and the alignment to the original translated exon are shown. The alignment score is given in per cent. The alternative exons are also marked in the Genomic DNA result view. In the Coding DNA and Translation result view the user can choose the alternative exons that should build the alternative coding DNAs or protein sequences. The results can be downloaded in several data formats. The YAML file contains all corresponding information and can later be uploaded and used for future analysis (290). Additionally, the results can be downloaded as General Feature Format (GFF) file (324). The figures can be downloaded in the Scalable Vector Graphics (SVG) format for further high quality processing (259). Example searches as well as further descriptions of the search parameters are provided on the help pages of WebScipio.
2.5.4 Results and Discussion
Identification of mutually exclusive spliced exons
The search for mutually exclusive spliced exons is based on three criteria: (1) The lengths of the mutually exclusive exons must be very similar, because these exons are supposed to code for the same part in the resulting protein structure, including identical secondary structural elements. (2) To be spliced in a mutually exclusive way, the exons must have similar splice sites and reading frames to be compatible with the previous and following exons. (3) The exons must encode homologous protein sequences, because their inclusion into the protein structure must be compatible with the corresponding local structural
environment. The search implemented in WebScipio is based on the availability of the gene structure. Firstly, mutually exclusive exon candidates are searched for using corresponding splice sites to the query exons and restricting the candidate length to similar reading frames (e.g. split codons in the query exon must result in split codons in the candidate exons). Total length difference is less restricted allowing length differences between query and candidate exons at the DNA-level in multiples of three for each additional or missing codon. These candidate exons are then filtered and scored based on the Blosum62 matrix. The best scoring, non-overlapping candidates are proposed to be alternative exons to the respective query exon, resulting in a cluster of mutually exclusive exons. With this approach, the absolute necessary constraints at the DNA-level that can be obtained by bioinformatics means are combined with biological information. Based on these criteria several cases can be distinguished: (A) alternative exons found in the surrounding introns of single internal exons should form true clusters of mutually exclusive exons, (B) alternative exons found for terminal exons most probably constitute multiple promoters or multiple poly(A) sites, (C) clusters of several exons in combination, which can be found by searching for candidates for all exons in all introns and up- and downstream regions, most probably represent cases of tandemly arrayed gene duplications or trans-spliced genes.
Figure 2.5-3: Example cases of mutually exclusive spliced exons, multiple promoters and multiple poly(A) sites. The figure illustrates three examples of genes containing mutually exclusive spliced exons, one example containing multiple promoters, and one containing multiple poly(A) sites. Dark grey bars and light grey bars mark exons and introns, respectively. The small blue bar represents an “intron?” that does not have canonical splice sites because an exon is
Figure 2.5-3: Example cases of mutually exclusive spliced exons, multiple promoters and multiple poly(A) sites. The figure illustrates three examples of genes containing mutually exclusive spliced exons, one example containing multiple promoters, and one containing multiple poly(A) sites. Dark grey bars and light grey bars mark exons and introns, respectively. The small blue bar represents an “intron?” that does not have canonical splice sites because an exon is